9 Sentences That Sound Like AI (and How to Rewrite Each One)
Readers can feel it before they can explain it. Here are nine patterns that make writing feel machine-made, each with a before and after.
You’ve probably had this experience. You open an article, read three sentences, and something feels off. Nothing is wrong exactly. The grammar is fine and the structure is tidy. But you can’t shake the feeling that no one in particular wrote it.
That feeling has a cause. Text from language models tends to share certain habits, and people who write a lot with AI help pick up the same habits without noticing. This article lists nine of them, with a before and after for each, so you can catch them in your own writing.
One caution first: none of these patterns prove anything. Plenty of humans write this way, and plenty of AI text avoids all of them. Treat this as an editing checklist, not a lie detector.
1. Every paragraph is the same size
The tell: Four neat sentences per paragraph, every paragraph, from start to finish. Real writing speeds up and slows down.
Before: Three medium paragraphs in a row, each ending on a mini-summary.
After: One long paragraph where you explain something carefully. Then a one-line paragraph that lands the point.
Fix: Read the shape of your page. If it looks like a wall of equal blocks, break one and merge two.
2. Everything comes in threes
The tell: “Faster, smarter, and more efficient.” Lists of three show up everywhere because they sound complete.
Before: “Our tool is simple, powerful, and flexible.”
After: “Setup took me ten minutes. The one thing I’d change is the export.”
Fix: Replace a trio with a single concrete detail, or two items if only two are true.
3. “It’s not X, it’s Y”
The tell: Constant contrast framing: “It’s not just a tool, it’s a movement.” One or two is fine. A dozen is a pattern.
Before: “This isn’t just a recipe. It’s a way of life.”
After: “This is the recipe I make when I have no time and nothing in the fridge.”
Fix: Say what the thing is. Drop the setup you’re knocking down.
4. No names, numbers or places
The tell: Vague abstractions: “many experts agree,” “studies show,” “in today’s fast-paced world.” Nothing you could check.
Before: “Many businesses struggle with social media.”
After: “The bakery down my street posted nine times in January and zero in February.”
Fix: Add one checkable detail per section: a name, a date, a number you actually know.
5. The throat-clearing opener
The tell: Openings that announce the topic instead of starting it: “In today’s digital age, social media has become increasingly important.”
Before: “In the world of productivity, time management is key.”
After: “I missed a deadline last Tuesday because I trusted a calendar I never opened.”
Get Kanisshka’s stories in your inbox
Join Medium for free to get updates from this writer.
Fix: Delete your first sentence. If the second sentence works as an opener, you’re done.
6. The tidy bow at the end
The tell: A conclusion that restates everything and finishes on a lesson: “In conclusion, it’s all about balance.” Every section ends with its own little moral.
Fix: End on the last useful thing, an open question, or a next step. You don’t need to summarize a 700-word article.
7. Stock words and phrases
The tell: Words that appear far more in model text than in everyday writing, such as “delve,” “landscape,” “tapestry,” “navigate,” “leverage,” “unlock.” The exact list shifts as models change, so don’t memorize one.
Before: “Let’s delve into the evolving landscape of remote work.”
After: “Here’s what changed when my team went remote.”
Fix: Whenever a word sounds like a press release, swap in the plain word you’d say out loud.
8. Confidence without risk
The tell: Everything is balanced and agreeable. No strong opinion, no admission of doubt, no “I was wrong.” Human writing takes small risks.
Before: “There are pros and cons to both approaches, and the best choice depends on your needs.”
After: “I’d pick the second one. The first looks better on paper and cost me a weekend.”
Fix: Pick a side where you honestly have one. Say what you’d do.
9. Invented specifics
The tell: The most serious one. AI text can state fake statistics, quotes, studies and links with complete confidence. A “personal story” with no checkable detail can also be made up.
Fix: Check every number, quote and link before publishing. If you can’t verify it, cut it or say it’s an estimate. Never present a made-up experience as yours.
Why none of this proves anything
It’s tempting to turn these into a test and accuse writers. Don’t. AI detection tools have a poor track record: OpenAI withdrew its own text classifier in 2023 citing low accuracy, and researchers at Stanford reported that popular detectors misclassified writing by non-native English speakers more often. Careful, formal human writing often trips every pattern above.
The reliable signals are about substance, not style. Are the facts checkable? Does the writer show their work? Do they take a position and stand behind it?
If you write with AI, here is a better workflow
- Start with your own notes, opinions and examples, then let the AI structure or tighten them.
- Put in the details only you know: names, numbers, mistakes, what you’d do differently.
- Cut anything you wouldn’t say out loud.
- Verify every claim and link.
- Read it once as a skeptic and delete the lines that sound like everyone else.
- Follow the rules of the platform you publish on, which may require telling readers when AI helped.
The short version
AI-sounding writing is mostly generic writing: even rhythm, vague claims, no risk. The cure is the same whether a human or a machine wrote the first draft. Add what only you know, say what you actually think, and check what you claim.
Which of these nine drives you up the wall the most? Tell me in the comments, and I’ll add the best ones to a follow-up. Follow me for more practical writing and AI guides.
Disclosure: I used AI to help draft this article and reviewed and edited it myself. Photos from Unsplash (free to use): Justin Morgan, Eugene Chystiakov and Fiona Murray-deGraaff.
